Welcome to Computational Musicology 2021!
Welcome to Computational Musicology 2021!
You can access the lecture notes at the following links.
- Week 6 · Fourier Analysis · Data Visualisation
- Week 7 · Audio Features · Dynamic Time Warping
- Week 8
- Week 9
- Week 10
- Week 11
This storyboard contains further examples from each week to inspire you. You can download the raw source code here.
What kind of portfolio do I need to create?
Your portfolio will be a 5–10-page dashboard in the style of the R package flexdashboard, using data from the Spotify API. Your dashboard should cover the following topics, but note that it is possible (and often desirable) for a single visualisation or tab to cover more than one topic.
- Background: Which tracks did you choose? Why? What questions will the dashboard answer?
- Track-level features: What insights into your corpus can be drawn from Spotify’s track-level features (valence, danceability, etc.)?
- Chroma features [pitch]: What insights into your corpus can be drawn from Spotify’s chroma features?
- Loudness [volume]: What insights into your corpus can be drawn from Spotify’s ‘loudness’ (power) features, either at the track level or the section level?
- Timbre features [timbre]: What insights into your corpus can be drawn from Spotify’s timbre features?
- Temporal features [duration]: What is notable, effective, or ineffective about structural segments, metre, rhythm, or beats in your corpus, as derived from Spotify features?
- Classification/regression: Where explicit labels in your corpus exist, how strong are Spotify features for classification or regression?
- Clustering: Where explicit labels in your corpus are lacking, which Spotify features generate potentially meaningful clusters?
- Contribution: What have you learned from this set of analyses? Who could these conclusions benefit, and how?
Depending on your topic, you may want to start with a text-based opening like this one; alternatively, you could put your most compelling visualisation directly on the first tab and just use the commentary to introduce your corpus and research questions.
The grading breakdown for the portfolio is as follows. The rubric was adapted from the Association of American Colleges and Universities (AAC&U) Inquiry and Analysis and Quantitative Literacy VALUE rubrics.
| Component | Points |
|---|---|
| Corpus selection | 6 |
| Assumptions | 6 |
| Representation | 6 |
| Interpretation | 6 |
| Analysis | 6 |
| Presentation | 6 |
| Transfer of learning | 6 |
Do the Dutch really listen to happier music? [track-level features]
In March 2019, The Economist published a graphic showing a worldwide dip in the emotional valence of the music to people people listen around February each year. The graphic also broke down the emotional valence for every month of the year for a selection of Spotify markets, and it revealed quite a surprise: although overall, there seemed to be two large groups, with Latin America and Spain listening to more positively valenced music than most of the rest of the world, the Netherlands stood on its own somewhere in between these two extremes. This graphic compares Spotify Top 50 in the Netherlands against Belgium, a more typical neighbouring country, as well as the worldwide average, on the standard valence–arousal model. More popular songs (worldwide) have larger dots; more danceable songs are yellow.
The Tallis Scholars sing Josquin more slowly than La Chapelle Royale – except for one part [chroma features].
This visualisation of two performances of the famous ‘Ave Maria’ setting of Josquin des Prez uses the Aitchison distance between chroma features to show how the two performances align with one another.
For the first four stanzas, the relationship between the performances is consistent: the Tallis Scholars sing the piece somewhat more slowly than La Chapelle Royale. For the fifth stanza (Ave vera virginitas, starting about 3:05 into the Tallis Scholars’ performance and 2:25 into La Chapelle Royale’s), the Tallis Scholars singing faster than La Chapelle Royale, but at the beginning of the sixth stanza (Ave preclara omnibus, starting about 3:40 into the the Tallis Scholars’ performance and 3:05 into La Chapelle Royale’s) the Tallis Scholars return to their regular tempo relationship with La Chapelle.
Although the interactive mouse-overs from ggplotly are very helpful for understanding heat maps, they are very computationally intensive. Chromagrams and similarity matrices are often better as static images, like the visualisation at left.